KG-Pipeline

Tooling and Interfacing for Knowledge Graphs Construction.

Structure

• core - core model of the framework
  • llm - LLM interfaces (OpenAi, Anthropic, Google)
  • provenance - TODO
  • metrics - metrics for evaluation
  • utils - helper methods
• tasks - single kgc tasks and tool interfaces
• datasets - input data acquisition and generation
• tests - tests with examples
• ui - web user intefaces
  • streamlit.py starts a simple web interface
• scripts - handy scripts for development

LLM for KG experiments

RML generation and RML mapping tests

Experiments for the paper "Towards self-configuring Knowledge Graph Construction Pipelines using LLMs - A Case Study with RML"

Iterate runs and requests LLM (generate RML+repair Turtle)

poetry run pytest -s kg_tests/test_final_experiment.py -k test_final 

Generate Stats for paper

poetry run pytest -s kg_tests/test_final_experiment.py -k test_final 

Inspect statistics and view F1 Scores for metrics

streamlit run kg_ui/execute.py

Final results https://akswnc7.informatik.uni-leipzig.de/~mhofer/paper_supplements/eswc24/kgc/

Config.yaml

see config.yaml

Install

pip install poetry
poetry install

Execute Tests

see tests

Install Remote Tools

All tools will be installed under tools dir.

RMLMapperJava

bash scripts/install-rmlmapperjava.sh

Development Notes

• Currently all resources for RML4LLM are located under llm4kg_tests/resources/